Yeah, I am going to say that you are going to be held back by having 4x2GB no matter what you try to do with voltage/RAM speed or timing/etc. I am not exactly an expert in the newer AMD processors (the vast majority of my AMD experience/expertise lies with the Athlon64 FX51 thru FX6x, A64 X2's, and then I switched to Intel almost exclusively), however I have done a fair amount of overclocking and benching with the X6 1100T, X4 960T, and X4 985BE. The motherboards I used were the CHV, Sabertooth 990FX, Gigabyte 990FX UD7, and the M4A99FX Pro, with the memory consisting of G.Skill Pi (1800 6-8-6-16), Trident X (1866 8-9-9-22),and Corsair Dominator GTX with theElpida Hypers (2200 8-9-8-24).
TThe best way I found to overclock the Phenom II Thubans is to go slowly, carefully, and use a pen and paper to keep track of every single change you make. The best performance I got was by overclocking the NB as high as possible (after finding the approximate max CPU speed via multiplier adjustments), relying on it to increase the processor speed as far as it will go, before touching multipliers. When you start to mess with the RAM, your speed settings will be determined by the NB speed, so having memory rated higher than what you intend to run it at is a great way to save a lot of headaches. The best memory for the platform are those that use the Elpida Hyper IC's but as you cannot buy them new anymore, you will have to track down a set (eBay or the OCN market are the absolute best places to get EOL/Discontinued RAM). It's very much worth the hassle, though, as these IC's can typically handle speed+timings combinations that no currently produced memory can touch! They are not the megahertz monsters some of the new Samsung and Hynix IC's are, but at any given speed they are almost guaranteed to run significantly tighter timings, I'm talking DDR3-2000 8-9-8 or better, 1866 6-8-6, and 1600 6-6-6!
I would start by pulling two of your DIMMs to increase the stability, and begin overclocking from there.
Start with everything at stock values, turn off turbo, and begin to increase the NB frequency by a few bumps, boot into Windows, run a quick SuperPI 16/32 run, and if it doesn't error out, reboot to BIOS and up it again. Continue until you get an error or fail to POST, at which point try increasing the voltage to the NB by 1-2 steps, until it will POST and complete the SuperPI run. Then continue upping the NB, testing, and increasing voltage until you either hit a wall or you reach the max voltage limit. Set it 3-10Mhz lower than the max stable speed to allow a bit of headroom/wiggleroom, and run Prime95 for an hour or so.
If all is good, then you can try increasing the CPU multiplier by 1 or 1/2 a step at a time. You will likely have already added some voltage to the CPU vCORE when upping the NB (unless you dropped the multiplier down to keep it as close to stock speed as possible just with a faster NB). Do pretty much the same thing you did with the NB in terms of small step, POST, and quick stress test (wPrime 1024m, Cinebench 11.5, or another heavily threaded CPU bench will do) before rebooting into BIOS and repeating, increasing the voltage when you get a failure to POST or error out on benches. Stop once you get to either "the wall", the max voltage, or most likely the temp limit. Once there, you can fine tune by trying to drop the voltage a little bit (working backwards with voltage one step @ a time), and once you find its minimum, set it 2-4 steps higher (but don't exceed where you were before trying to lower it).
Using the method above, which I wrote in an abbreviated format (I'm typing this on my phone), I have gotten the 1100T to 4.34Ghz (Water-cooled, w GTX580 Classy Ultra 3GB in the loop and a HWL SR1 360 + MagiCool 240 rad combo), the 985BE to 4.4Ghz (on air, using the Thermalright SA SB-E with 3x Sanyo Denkei San Ace 120x38mm 4200rpm/179CFM/22.8mmH2O fans, lapped CPU + HSF with 300 to 8000grit sandpaper, and "tinted" the IHS and HSF's base with PK1), and the 960T both locked/unlocked to 4.45Ghz/4.1Ghz (H2O, Raystorm block, GTX580 3GB + MB chipset/VRMs in loop, with 2x HWL GTX 480 rads using Push-Pull 120x38mm 3100rpm/143cfm/12.7mmH2O fans).
When it comes to memory, you have both the NB frequency AND the RAM multiplier to use for adjusting the speed, which is why we want to leave 3-10Mhz of wiggleroom with the NB clock.
Then, because nothing in life is easy, you have the memory timings, which are critical to performance and vary considerably depending on the memory speed, type of IC's, voltage, number of ranks, whether you're using single or double sided DIMMs, the density of the IC's, and the number of DIMMs in use, among many other variables.
With the Phenom II, you have a chip without the strongest IMC (although Thubans are better than Deneb IME), so don't expect huge frequency bumps. Instead, you are wanting to get the absolute tightest timings possible at the highest stable speed. With the Phenom II, I would take something like DDR3-1646 6-8-7-19 1T over DDR3-1884 9-10-9-24 every time. Basically, if you up the memory multiplier by one and have to increase any of the first three timings by more than one, you are not gaining performance and may actually be losing some.
While some people argue otherwise, it's absolutely true that AMD prefer tight timings with average speeds, and Intel wants huge speeds even with loose timings. It's just how it is.
When tightening timings, start with CAS (1st), and if it is good with a single drop, move to the third, then the fourth, then the second. Every time you drop each timing, you need to boot/POST and then run SuperPI 32M, and if it passes record the time, and then run MaxxMEM and record the three bandwidth numbers and the latency. Do this every time, for each individual timing change!
If you start to see the bandwidth numbers go down, or latency up, or SuperPI is taking more than 1-2sec extra, you have changed something and thrown the system out of balance. Also, if you fail to post or fail SP 32M, you need to supply more memory voltage (I would not get above 1.675v just for peace of mind) or bump the timing back up.
That's my recommendation. Hope it helps, and sorry I can't flesh it out further as it's been a while since I have benched one of these, not to mention I'm writing from my phone!